Classification of Humans and Animals from Radar Signals using Multi-Input Mixed Data Model

Abstract

Radars have been around us for a long time, about a century indeed. We have witnessed their usage in World War II for locating airplanes. In today's world, radars are used in aviation, defence and also in automobiles. Although highly accurate radars have limited classification capabilities, most of the common radars existing today cannot classify and label an object correctly. For instance, if a private jet and a fighter jet travel at similar speeds and similar altitudes, radar cannot always report the difference between them. We concentrated on Ground Surveillance Radars (GSRs) for our study. The differentiation of humans and animals by GSR is one of the most difficult tasks to discern based on radar signals, primarily because of their comparable sizes, speeds and forms of motion. Multiple cameras are required in the current scenario to display the footage, and a lot of manual work is done to identify an animal and an individual from the footage. When performed manually, this classification is a tedious task and is highly dependant on the camera's visibility, which is often inefficient. To overcome this manual task, a novel concatenated convolutional neural network (CNN) model has been proposed that takes the I/Q matrix (radar signal data) and the geolocation type as its input and performs a binary classification to categorize animals and humans. The proposed model has achieved an (Area under the Receiver Operating Characterstic curve(AU-ROC) of 99%. © 2021 IEEE.